Recrystallization, which can occur dynamically or statically, is an important phenomenon causing microstructure changes in deformed metals and therefore affecting the properties of the material. Even though extensive investigations have been carried out on the numerical modeling of recrystallization, the literature lacks accurate recrystallization models which are able to predict microstructure evolution under multi-pass conditions. Although some efforts have been reported in this direction, most of them either lack experimental validation or only provide qualitative agreement in selected deformation conditions. The connections between static recrystallization (SRX), dynamic recrystallization (DRX), post-dynamic recrystallization (PDRX) and grain growth (GG) are usually oversimplified. Furthermore, most of them are not designed for variable thermal and/or mechanical conditions and are therefore difficult to use for industrial applications. In this work a physically-based two-site mean field model has been developed to describe the microstructural evolution of 304L austenitic stainless steels. The originality of the model lies in: (a) the interaction of each representative grain with two homogeneous equivalent media, with high and low dislocation density, respectively; (b) the relative weight of the two media is functionally related to their volume fractions; (c) nucleation and disappearance of grains make the data structure variable in time; (d) the model parameters vary with temperature and strain rate but do not depend on grain size in DRX conditions, and become only temperature dependent in static conditions (SRX/PDRX/GG); (e) quantitative agreement with experimental results is obtained in terms of (i) recrystallization kinetics, (ii) stress-strain curves, (iii) recrystallized grain size, and (f) it is designed to be used in multi-pass conditions, with variable temperature and strain rate. To verify and validate the model, torsion tests were conducted over a wide range of conditions for investigation of DRX. Subsequent annealing after termination of deformation, which led to SRX or PDRX depending on the applied strain, was also carried out. The model parameters were first estimated from experimental and literature data, and were further tuned by inverse analysis. It is found that all identified model parameters evolve with temperature and strain rate in a physically consistent way. The application of this proposed model to DRX, SRX/PDRX/GG is then analyzed, taking into account the effects of deformation temperature, strain rate, applied strain, as well as initial grain size. Good quantitative agreements with measured data are obtained in the different recrystallization regimes, which opens the possibility of modeling multi-pass operations compatible with industrial applications. A few in situ heating experiments were carried out to provide a better understanding of the SRX/PDRX/GG mechanisms. The role of annealing twins is tentatively discussed: it seems to promote both nucleation and grain boundary migration.